Π peb tham txog lub sij hawm series forecasting. Ib qho laj thawj txuas ntxiv yuav yog ib tsab xov xwm ntawm kev txheeb xyuas qhov tsis txaus ntseeg.
Daim ntawv thov
Anomaly detection yog siv rau hauv thaj chaw xws li:
1) Kev kwv yees ntawm cov khoom tawg
Yog li, xyoo 2010, Iranian centrifuges tau tawm tsam los ntawm tus kab mob Stuxnet, uas tau teeb tsa cov cuab yeej siv rau kev ua haujlwm tsis zoo thiab cuam tshuam qee cov khoom siv vim kev hnav nrawm dua.
Yog tias qhov kev kuaj pom tsis zoo tau siv los ntawm cov khoom siv, qhov xwm txheej tsis ua haujlwm yuav raug zam.
Kev tshawb nrhiav kev tsis sib haum xeeb hauv kev ua haujlwm ntawm cov cuab yeej siv tsis yog hauv kev lag luam nuclear xwb, tab sis kuj tseem nyob hauv metallurgy thiab kev ua haujlwm ntawm aircraft turbines. Thiab nyob rau hauv lwm qhov chaw uas siv cov kev kuaj mob uas kwv yees tau pheej yig dua li qhov ua tau poob vim qhov kev puas tsuaj tsis tuaj yeem tshwm sim.
2) Kev twv ua ntej kev dag
Yog tias cov nyiaj rho tawm ntawm daim npav koj siv hauv Podolsk hauv Albania, cov kev hloov pauv yuav tsum tau kuaj xyuas ntxiv.
3) Kev txheeb xyuas qhov txawv txav ntawm cov neeg siv khoom
Yog tias qee cov neeg siv khoom pom tus cwj pwm txawv txav, tej zaum yuav muaj teeb meem uas koj tsis paub txog.
4) Kev txheeb xyuas qhov txawv txav xav tau thiab thauj khoom
Yog tias kev muag khoom hauv lub khw FMCG tau poob qis dua qhov kev ntseeg siab ntawm qhov kev kwv yees, nws tsim nyog nrhiav qhov laj thawj rau qhov tshwm sim.
Txoj hauv kev los txheeb xyuas qhov tsis txaus ntseeg
1) Txhawb Vector Tshuab nrog Ib Chav Kawm Ib-Class SVM
Tsim nyog thaum cov ntaub ntawv hauv cov txheej txheem kev cob qhia ua raws li kev faib tawm ib txwm, tab sis cov txheej txheem xeem muaj qhov tsis zoo.
Ib-chav txhawb nqa vector tshuab tsim ib qho chaw tsis muaj kab nyob ib ncig ntawm keeb kwm. Nws muaj peev xwm teem caij txiav tawm rau cov ntaub ntawv uas suav tias yog qhov tsis zoo.
Raws li kev paub dhau los ntawm peb pab pawg DATA4, One-Class SVM yog qhov feem ntau siv algorithm los daws qhov teeb meem ntawm kev nrhiav qhov tsis sib xws.
2) Txoj kev cais hav zoov
Nrog rau txoj kev "random" ntawm kev tsim cov ntoo, emissions yuav nkag mus rau hauv nplooj thaum ntxov (ntawm qhov tob tob ntawm tsob ntoo), i.e. emissions yooj yim dua rau "kev cais tawm." Kev rho tawm ntawm cov txiaj ntsig tsis zoo tshwm sim hauv thawj iterations ntawm algorithm.
3) Elliptic hnab ntawv thiab cov txheej txheem kev txheeb cais
Siv thaum cov ntaub ntawv ib txwm faib. Qhov ze ntawm qhov ntsuas yog mus rau tus Tsov tus tw ntawm qhov sib tov ntawm kev faib tawm, qhov ntau anomalous tus nqi.
Lwm txoj hauv kev txheeb cais kuj tuaj yeem suav nrog hauv chav kawm no.
Duab los ntawm dyakonov.org
4) Metric txoj kev
Cov txheej txheem muaj xws li algorithms xws li k-ze neeg nyob ze, k-ze neeg nyob ze, ABOD (lub kaum ntse ntse outlier detection) los yog LOF (lub zos outlier factor).
Tsim nyog yog tias qhov kev ncua deb ntawm qhov tseem ceeb hauv cov yam ntxwv sib npaug los yog normalized (kom tsis txhob ntsuas tus boa constrictor hauv parrots).
Lub k-ze ze ze cov neeg zej zog algorithm xav tias ib txwm muaj nuj nqis nyob hauv ib cheeb tsam ntawm ntau qhov chaw, thiab qhov kev ncua deb mus rau qhov tsis sib xws yuav ntau dua li qhov sib cais hyperplane.
5) Cov txheej txheem ua pawg
Lub ntsiab lus ntawm pawg txheej txheem yog tias yog tias tus nqi ntau dua li ib qho nyiaj deb ntawm pawg chaw, tus nqi tuaj yeem suav tias yog qhov tsis zoo.
Qhov tseem ceeb tshaj plaws yog siv cov algorithm uas yog pawg cov ntaub ntawv, uas yog nyob ntawm cov haujlwm tshwj xeeb.
6) Cov txheej txheem tseem ceeb
Tsim nyog qhov twg cov lus qhia ntawm kev hloov pauv loj tshaj plaws nyob rau hauv dispersion yog highlighted.
7) algorithms raws li lub sij hawm series forecasting
Lub tswv yim yog tias yog tias tus nqi poob sab nraum qhov kev kwv yees kev ntseeg siab lub sijhawm, tus nqi raug suav tias yog qhov tsis zoo. Txhawm rau kwv yees lub sijhawm series, algorithms xws li triple smoothing, S (ARIMA), boosting, thiab lwm yam yog siv.
Lub sij hawm series forecasting algorithms tau tham nyob rau hauv tsab xov xwm dhau los.
8) Kev saib xyuas kev kawm (regression, classification)
Yog hais tias cov ntaub ntawv tso cai, peb siv algorithms xws li linear regression mus rau recurrent networks. Cia peb ntsuas qhov sib txawv ntawm qhov kwv yees thiab tus nqi tiag tiag, thiab kos cov lus xaus rau qhov twg cov ntaub ntawv deviates los ntawm tus qauv. Nws yog ib qho tseem ceeb uas lub algorithm muaj peev xwm generalization txaus thiab hais tias cov txheej txheem kev cob qhia tsis muaj anomalous qhov tseem ceeb.
9) Cov qauv xeem
Cia peb mus txog qhov teeb meem ntawm kev tshawb nrhiav qhov tsis txaus ntseeg raws li qhov teeb meem ntawm kev tshawb nrhiav cov lus pom zoo. Cia peb decompose peb feature matrix siv SVD los yog factorization tshuab, thiab coj qhov tseem ceeb nyob rau hauv lub tshiab matrix uas yog ho txawv los ntawm cov thawj sawv daws yuav anomalous.
Duab los ntawm dyakonov.org
xaus
Hauv tsab xov xwm no, peb tau tshuaj xyuas cov txheej txheem tseem ceeb rau kev kuaj pom tsis meej.
Nrhiav kev tsis txaus ntseeg tuaj yeem raug hu ua kos duab. Tsis muaj qhov zoo tshaj plaws algorithm lossis mus kom ze, kev siv uas daws txhua yam teeb meem. Feem ntau cov txheej txheem yog siv los daws qhov teeb meem tshwj xeeb. Kev kuaj pom tsis zoo yog ua los ntawm kev siv ib-chav txhawb nqa vector tshuab, cais hav zoov, metric thiab pawg txheej txheem, nrog rau kev siv cov khoom tseem ceeb thiab lub sijhawm kwv yees.
Yog tias koj paub lwm txoj hauv kev, sau txog lawv hauv cov lus rau kab lus.
Tau qhov twg los: www.hab.com